On 24 July 2026, a single judge of the Delhi High Court, Justice Amit Bansal, delivered a judgment[1] on whether training a large language model (LLM) on copyrighted content results in copyright infringement. The judgment, delivered after 32 hearings, dismissed ANI’s application demanding an interim injunction, holding that OpenAI’s storage of ANI’s literary works and copyrighted content for training falls within the ambit of Section 52(1)(a) of the Copyright Act, 1957, and that the outputs given by ChatGPT were not the same or substantially similar to ANI’s news content.
In November 2024, ANI, India’s most prominent news agency, filed a suit against OpenAI. This case becomes significant as it is the first such case to be filed in India. The two key issues placed before the court were: (i) that OpenAI had stored ANI’s copyrighted news content to train the LLMs underlying ChatGPT, and (ii) that ChatGPT reproduced ANI’s content in the output it generated for users.
OpenAI argued that its models are not trained in India and its servers are located in the United States. The defendants also pointed out that a copyright claim cannot be made on factual information, that machine learning is transformative and does not copy original content, and that publishers who are not willing to be crawled can exclude themselves through blocklists.
Since the issues posed before the court were novel, the Court appointed two amici curiae, namely Adv. Adarsh Ramanujan and Prof. (Dr.) Arul George Scaria, and along with these the Court also permitted intervention by industry bodies representing publishers, digital news publishers, music producers, AI developers and tech-policy groups. The judgment was reserved on 27 March, and the final four issues framed were: (i) whether Indian courts have territorial jurisdiction given that the defendant’s servers are in the United States; (ii) whether storage of ANI’s data for training results in copyright infringement; (iii) whether use of that data to generate responses for users is infringement; and (iv) whether such use is protected by Section 52 of the Copyright Act, 1957.
The issue of jurisdiction was the only one that was decided in favour of ANI. The Court declined to separate the issue of AI training and that of AI-generated outputs. It said that training an AI and the responses it produces are closely interconnected. Since the AI’s responses are available and can be accessed in India, the Court held that it cannot dismiss the case simply because OpenAI’s servers are located in the United States and not in India. Thus, the Court clarified that foreign-based AI companies can still be sued in India if the AI and its content are available in India.
However, the Court rejected ANI’s claim regarding AI-generated responses not because such claims are legally impossible but because there was a lack of evidence in this regard. The examples submitted by ANI in the court were published after the AI models had completed their training (April 2022 for GPT-4 and April 2024 for GPT-4o). The Court found that such responses were most likely generated through Retrieval-Augmented Generation (RAG), where the AI draws on information directly from ANI’s live website, rather than taking it from its training data. Since ANI could not show that the AI had reproduced its copyrighted content in a substantially similar way, the claim could not be sustained.
Additionally, the Court left one of the issues unanswered, i.e. if the AI copies content word for word from a live website, will that result in copyright infringement?
The most debatable part of the judgment was regarding Section 52(1)(a) of the Copyright Act. The Court used a two-pronged test:
- Does AI training fall under one of the purposes mentioned in Section 52(1)(a) of the Copyright Act?
- If yes, is the use fair dealing?
The Court held that commercial use does not automatically exclude fair dealing. It explained that whenever lawmakers wanted an exception to apply only to non-commercial use, they clearly said so. Since Section 52(1)(a) does not contain such a restriction, commercial AI training can qualify.
The Court also added that AI training is a form of research as it involves collecting data and converting it into a machine-readable format. The Court held that even though Section 52 was last amended in 2012, it should be interpreted in a way that keeps up with new and upcoming technology like AI.
However, given this rationale, the Court’s reasoning can be challenged because the fair use doctrine in the U.S. is broad, but Indian copyright law maintains a limited list of fair dealing exceptions, and AI training is not expressly included. By interpreting the term “research” to cover commercial AI training, the Court expanded Section 52(1)(a) beyond its original scope of protecting academic research. Lastly, the Court also refused to grant an injunction, holding that ANI failed to prove financial loss, could have blocked OpenAI’s web crawlers, had offered to license its content, and that restricting AI training would slow innovation and AI research and development in India. However, it is pertinent to note that publishers cannot prove actual financial losses while the AI licensing market is very new and evolving, and that merely giving a licence does not amount to giving consent to unauthorised use, as copyright grants the owner the exclusive right to control the usage of their work.
[1] ANI Media Pvt. Ltd. v. OpenAI OpCo LLC Delhi High Court, CS(COMM) 1028/2024, Decided on 24 July 2026
O.P. Jindal Global University
